--- title: JPDC 2024 Digest venue: JPDC year: 2024 date: '2024-01-01' tags: [] paper_count: 12 draft: false --- 12 papers selected. --- ### Read/write fence-free work-stealing with multiplicity *Armando Castañeda, Miguel Piña* **TL;DR** — Presents a work-stealing deque algorithm that eliminates read/write memory fences while tolerating multiplicity, achieving provably correct concurrent access without costly barriers. **Why notable** — Advances the theoretical foundations of lock-free scheduler data structures by decoupling correctness from fence instructions, directly impacting runtime system design. --- ### Reliable communication in dynamic networks with locally bounded byzantine faults *Silvia Bonomi, Giovanni Farina, Sébastien Tixeuil* **TL;DR** — Characterizes the conditions under which reliable broadcast is achievable in dynamic networks where Byzantine faults are locally bounded rather than globally counted. **Why notable** — Provides tight impossibility and achievability results for a practically motivated fault model, extending classical Byzantine agreement theory to time-varying topologies. --- ### Local certification of graph decompositions and applications to minor-free classes *Nicolas Bousquet 0001, Laurent Feuilloley, Théo Pierron* **TL;DR** — Develops local distributed certification schemes for graph decompositions, proving tight certificate-size bounds for minor-free graph families. **Why notable** — Connects structural graph theory to distributed verification, giving new tools for designing space-efficient proof-labeling schemes in anonymous networks. --- ### Eventually lattice-linear algorithms *Arya Tanmay Gupta, Sandeep S. Kulkarni* **TL;DR** — Introduces the eventually lattice-linear class of distributed algorithms that converge to a lattice-linear fixed point, enabling new self-stabilization constructions. **Why notable** — Unifies several stabilizing algorithm families under a clean algebraic framework that simplifies correctness proofs and inspires new design patterns. --- ### Construction algorithms of fault-tolerant paths and disjoint paths in k-ary n-cube networks *Mengjie Lv, Jianxi Fan, Baolei Cheng, Jia Yu 0003 *et al.** **TL;DR** — Proposes efficient algorithms to construct maximally fault-tolerant Hamiltonian paths and node-disjoint paths in k-ary n-cube interconnection networks under edge and vertex failures. **Why notable** — Yields concrete routing strategies for torus-based HPC fabrics that remain functional under high fault counts, with tight proofs of optimality. --- ### MapReduce algorithms for robust center-based clustering in doubling metrics *Enrico Dandolo, Alessio Mazzetto, Andrea Pietracaprina, Geppino Pucci* **TL;DR** — Designs MapReduce algorithms for k-median and k-means clustering with provable approximation ratios that are robust to outliers in doubling metric spaces. **Why notable** — Delivers the first round-efficient MapReduce clustering algorithms with simultaneous robustness and approximation guarantees grounded in metric space theory. --- ### DuMato: An efficient warp-centric subgraph enumeration system for GPU *Samuel Ferraz, Vinícius Vitor dos Santos Dias, Carlos H. C. Teixeira, Srinivasan Parthasarathy 0001 *et al.** **TL;DR** — Presents a warp-centric GPU programming model for subgraph enumeration that eliminates load imbalance through dynamic work redistribution across warps. **Why notable** — Achieves orders-of-magnitude speedups over CPU baselines on graph pattern mining by rethinking how irregular workloads are mapped onto SIMT hardware. --- ### An efficient sequential consistency implementation with dynamic race detection for GPUs *Abdulaziz Tabbakh, Murali Annavaram* **TL;DR** — Implements sequential consistency on GPUs via a dynamic race detector that inserts fences only where data races are actually detected at runtime. **Why notable** — Shows that strong memory model guarantees on GPUs need not incur pervasive overhead, opening a path toward safer GPU programming models without sacrificing performance. --- ### General-purpose data stream processing on heterogeneous architectures with WindFlow *Gabriele Mencagli, Massimo Torquati, Dalvan Griebler, Alessandra Fais *et al.** **TL;DR** — Extends the WindFlow library with a unified programming model and runtime that transparently targets CPUs, GPUs, and FPGAs for streaming dataflow applications. **Why notable** — Demonstrates practical performance portability for stream processing across radically different hardware, with a clean operator model grounded in parallel patterns theory. --- ### Revisiting I/O bandwidth-sharing strategies for HPC applications *Anne Benoit, Thomas Hérault, Lucas Perotin, Yves Robert *et al.** **TL;DR** — Analyzes optimal I/O bandwidth allocation policies for HPC checkpointing and data-intensive applications under shared storage systems, deriving scheduling algorithms with provable performance bounds. **Why notable** — Provides theoretically grounded guidance for I/O resource management at scale, a pervasive bottleneck in modern HPC deployments. --- ### Scalable atomic broadcast: A leaderless hierarchical algorithm *Lucas V. Ruchel, Edson Tavares de Camargo, Luiz Antonio Rodrigues, Rogério C. Turchetti *et al.** **TL;DR** — Proposes a leaderless hierarchical atomic broadcast protocol that eliminates single-leader bottlenecks and scales to large distributed systems with reduced message complexity. **Why notable** — Advances Byzantine-fault-tolerant broadcast by removing the leader as a scalability and availability bottleneck while preserving total-order guarantees. --- ### Stab-FD: A cooperative and adaptive failure detector for wide area networks *Pierre Sens 0001, Luciana Arantes, Anubis Graciela de Moraes Rossetto, Olivier Marin* **TL;DR** — Presents a stabilizing failure detector that cooperatively adapts its timeout parameters across WAN nodes to achieve eventual accuracy under varying latencies. **Why notable** — Tackles the long-standing challenge of practical failure detection in geo-distributed systems with a rigorous self-stabilizing design that avoids static parameter tuning.